Arizona is known for a lot of things: the Grand Canyon, SB1070, soaring summertime temperatures. But robust political polling isn't necessarily one of them.
That said, when stat hound Nate Silver of FiveThirtyEight predicts something, folks take note.
According to the FiveThirtyEight 2016 General Election Forecast unveiled this week, Donald Trump has a 50.5 percent chance of winning Arizona, compared to Hillary Clinton's 49.3 percent. That's a surprisingly slim margin, leading Silver to place Arizona in his "States to watch" category.
For the sake of comparison, the Cook Political Report classifies Arizona as "leans Republican" — one step above a political "toss-up," but not "likely Republican" or "solidly Republican."
(Silver's national predictions give Trump a 20 percent chance of choosing the White House drapes come January.)
If the election were held today, Silver's crystal ball shines a little more clearly, giving Trump a 65.4 percent chance of winning to Clinton's 34.5 percent in his "Now-cast." The drastic difference in outcomes has to do with the particular statistical models used: The computer model "weights recent polls more heavily and is more aggressive in calculating a trend line," writes Silver.
This matters, he explains, because "fluctuations in the polls in the summer are often statistical noise or short-term bounces," and so the model predictions for November are "trained to be conservative in reacting to them." The later in the race, the more meaningful fluctuations become, and so if you tell the computer the election is being held today, it will take those fluctuations more seriously.
In terms of an the actual vote tally, FiveThirtyEight sees Trump currently leading Clinton by a scant margin of 45.0 percent to 44.9 percent in Arizona, with another 8.7 percent accounted for by Libertarian Party candidate Gary Johnson.
Political polls are always a tricky thing, often said to reflect the political landscape at a particular moment in time – each party tends to see a spike in popularity when it holds its political convention, for instance.
What makes Silver's predictions stand out has to do with the complex models he and his team use to assess, weigh, and combine the results from all available polls in a given state. For instance, polls that have a track record of being more accurate are given more weight, and the tendency of some polls to lean left or right is taken into account in the final model. Then there are the demographic, historical, and other complicated statistical models FiveThirtyEight bakes into its formula.
The complex methodology catapulted Silver to forecasting fame in the 2008 presidential race, when his FiveThirtyEight.com blog accurately predicted the electoral outcome in 49 of the 50 states. (Silver had already made a name for himself in baseball's sabermetrics community, having developed a forecasting system called PECOTA in the early 2000s.) The 2012 presidential race solidified Silver's place in political statistical history, when he went a perfect 50 for 50. By then he had hitched FiveThirtyEight's wagon to the New York Times; in 2014, Silver relaunched under the aegis of ESPN, broadening his scope from politics to data journalism in general.
If you like this story, consider signing up for our email newsletters.
SHOW ME HOW
But also, Trump's really underperforming in red states. Winning Texas by 7% instead of 17%, for instance. That impacts the national numbers.— Nate Silver (@NateSilver538) June 29, 2016
Political geeks have been anxiously awaiting Silver's 2016 General Election Forecast. And the debut of this year's model didn't disappoint, equipped as it is with a new statistical feature: "Polls-plus" data, which adds economic data and voting history (among other things) to the "Polls-only" forecast.
The Polls-plus model for Arizona shows Trump leading by pretty much the same margin as the Now-cast, 65.4 to 34.6.
Confused? Fascinated? Scared? Thirty-three percent of all three? Stay tuned. Silver's most accurate prediction right now is that his model is very likely to look different a month, two months, and three months down the road to November.